The last of the three. We think this is the most important one.
A comparison of how clever they are dates within three months. Client information that has gone out cannot be taken back.
Personal plans and business plans are treated differently
Whether the text you put in is used to train the AI depends on the plan. We checked what each company states officially (as at 16 August 2026).
| Used for training? | |
|---|---|
| ChatGPT (business: Business / Enterprise / API) | States explicitly that “by default, we do not use your business data to train our models” |
| ChatGPT (personal: Free / Go / Plus / Pro) | Used for training. An opt-out — a setting that stops it — is provided |
| Claude (Team / Enterprise) | States explicitly that “by default, we do not train our models on your content” |
| Claude (Free / Pro / Max) | An opt-out is provided |
| Gemini (Google Workspace) | States explicitly that “your content is not used to train models outside your organisation without permission” and that it is “not reviewed by humans” |
The sources are listed at the end of the article.
What that tells you
What all three have in common is that the business plans say “not used for training by default”. Put the other way round, it also means that on a personal plan, that is not the case unless you change the setting.
So the decision is fairly simple, we think.
If company information is going in, be on a business plan.
The difference is a few thousand yen a month. For a company handling clients’ property information, that is not the place to save money.
Even so, do not rely on the plan alone
Being on a business plan does not mean anything can go in.
- Settings can be changed. Somebody may change them
- Plans change along the way. Someone will carry on using a personal one
- Company information that never goes in cannot leak
So alongside the plan, we set rules inside the company.
Our rules
Nothing complicated. Four of them.
1. No property names, addresses or client names
Instead of “Unit 302, XX Mansion” we write “a unit in a rental block” and ask that. The AI usually does not need that information to answer.
2. Drawings and contracts do not get uploaded as they are
We type out only the part we need and paste that. It takes longer, but it keeps us in a state where we know what we handed over.
3. What is used for work runs on the company’s plan
Meaning: no company work on a personal account. It is also to avoid company information sitting inside the account of someone who has left.
4. Whatever comes out is checked by a person
At the bottom of these OS Notes articles it says that the article was drafted by AI and checked by our staff. If the check is going to be skipped, better not to publish at all.
These rules are nothing as impressive as what a large company would have. They are what a company with few people settled on, within what it can remember. Even so, it is much better than nothing.
Pricing comes in three shapes
We are not writing figures. By the time you read this they have probably changed. (For the same reason, we do not publish a price list for our works either.)
Instead, here are the shapes. Those will not change for a while.
1. A monthly fee for one person A flat rate for one. Start here when you are trying it out. If company information is going in, do not stop here.
2. A monthly fee per person You pay for as many people as use it. An admin screen comes with it, and “not used for training by default” usually starts here. There can be a minimum number of people (Claude’s Team plan, for example, starts at two).
3. Pay for what you use (API) The shape for building it into your own app. Use none and you pay nothing, but without a cap there is no ceiling. Our business app uses this shape, and we put the mechanism that stops runaway usage in first.
The costs that get missed
Two of them never appear on a price list.
The time it takes to learn. The time between handing it to someone and it sticking costs far more than the monthly fee.
The time it takes to switch. Relearning how to ask so that it does what you meant. We wrote about this in the second article too, and we think it is the biggest one.
What to check before you sign (summary)
When someone in the trade asks, this is the order we go through.
- Have you decided what it is for? If not, there is no need to sign up yet (the first article)
- Is company information going in? If it is, a business plan is the only option
- Have you tested it on your own real documents? Not comparison articles — your own paperwork (the second article)
- Have you decided what will not go in? The habit inside the company comes before the plan
- Have you decided who checks? Nothing goes out as it came
Numbers 1 and 4 can be decided before you sign anything. And our experience so far is that bringing AI in without those two usually does not work.
Finally
We have received nothing from any of the companies named in this series. We have not ranked them either.
We wrote it because we thought contractors of about our size would get stuck in the same places. If it gives you something to decide with, that is enough.
If you have a “here is what we decided” of your own, send it through the box below. The harder something is to hear, the more it is worth publishing.
Sources (all checked on 16 August 2026)
- OpenAI — Enterprise privacy: “by default, we do not use your business data to train our models”
- Anthropic — Claude pricing plans: Team and Enterprise “do not train models on your content by default”
- Google Workspace — Generative AI privacy: “your content is not used to train generative AI models outside your organisation without permission, and it is not reviewed by humans”
Prices and policies change. Please check each company’s own pages before you sign up.
This series
- Decide what not to hand over, before you pick one
- Using Claude, ChatGPT and Gemini for different jobs
- What happens to what you put in — things to check before you sign (this article)